Identification of Tree Locations in Geographic Images

treeid1By: David Brown

This project describes an interactive image analysis algorithm for the correct placement of trees within a virtual reality wild re visualization system. The wild re visualization system is VFIRE (Virtual Fire in Realistic Environments), which uses fire simulation results produced by software programs such as FARSITE to place the user into an immersive, arti cial environment where he or she can see graphical depictions of real or fictitious wild fire events. The purpose of VFIRE is to assist in training and planning for actual fires, and our purpose is to place trees within the arti cial environment at locations corresponding to their locations in the real environment. The goal was to achieve adequate tree-placement accuracy using any, arbitrarily chosen image, assuming no constraints on the minimum suitability of that image for tree detection. treeid2This goal led to the selection of a template-matching algorithm wherein the user selects a particular tree as a template and begins to search the image for similar-looking trees. Tree detection results are presented for a 1-meter panchromatic image of the current area of interest, Kyle Canyon, in Southern Nevada. It is difficult to perceive the aesthetic effect of tree-detection inaccuracies in terms of numeric error rates. Therefore, screen shots are provided representing tree-detection results placed into three broad, subjective categories: good, adequate, and poor. The results obtained were found to be roughly fifteen percent good, seventy percent adequate, and fifteen percent poor, where accuracy was directly related to the suitability of the terrain for tree detection in each particular region of the image.treeid3
Brown DT. 2008. Identi cation of Tree Locations in Geographic ImageComputer Science and Engineering. MS CS:73.